controller gates logs and metrics on the hive capability and mirrors the grant
This commit is contained in:
parent
3b1d118f25
commit
e7e3281098
1 changed files with 519 additions and 38 deletions
|
|
@ -23,11 +23,13 @@ use problem_details::ProblemDetails;
|
|||
use serde::Serialize;
|
||||
use swarm_controller_api::permissions::Permissions;
|
||||
use swarm_queue_client::agent_permissions;
|
||||
use swarm_secret_client::observability::Grant;
|
||||
use swarm_wire::agent_permissions::{
|
||||
AgentPermissions, Import, ImportReply, hive_from_import_subject, import_subject_wildcard,
|
||||
};
|
||||
use utoipa::ToSchema;
|
||||
|
||||
use crate::observability_grant::{self, grant_of, mirror, require_tolerant_hive};
|
||||
use crate::routes::{AppState, error_problem, wanted_error_status};
|
||||
use crate::schedules::{live, queue, read_or_default, refused, write, write_revision};
|
||||
use crate::wanted::WantedWriter;
|
||||
|
|
@ -102,6 +104,31 @@ fn logged(op: &'static str, agent: &str, problem: ProblemDetails) -> ProblemDeta
|
|||
problem
|
||||
}
|
||||
|
||||
/// Bring the store's mirror of `agent`'s `logs`/`metrics` grant in line with
|
||||
/// a record that went from `before` to `after`, after the record is written.
|
||||
/// A failure is an error to the caller, who repeats the request: `mirror`
|
||||
/// rewrites whenever either record holds a grant.
|
||||
async fn mirrored(
|
||||
agent: &str,
|
||||
before: &AgentPermissions,
|
||||
after: &AgentPermissions,
|
||||
) -> Result<(), ProblemDetails> {
|
||||
let held = |v: &AgentPermissions| v.permissions.clone().unwrap_or_default();
|
||||
mirror(&observability_grant::Store, agent, &held(before), &held(after))
|
||||
.await
|
||||
.map_err(|e| {
|
||||
tracing::warn!(agent, error = %format!("{e:#}"), "agent permissions: mirroring the grant failed");
|
||||
error_problem(
|
||||
StatusCode::INTERNAL_SERVER_ERROR,
|
||||
&format!("the permissions are stored, but mirroring logs/metrics to the secret store failed: {e:#}"),
|
||||
)
|
||||
})
|
||||
}
|
||||
|
||||
fn hive_names(state: &AppState) -> Vec<String> {
|
||||
state.hives.iter().map(|h| h.name.clone()).collect()
|
||||
}
|
||||
|
||||
fn queue_problem(e: &anyhow::Error) -> ProblemDetails {
|
||||
tracing::warn!(error = %format!("{e:#}"), "agent permissions: queue failure");
|
||||
error_problem(wanted_error_status(e), &format!("{e:#}"))
|
||||
|
|
@ -185,6 +212,19 @@ pub(crate) async fn put_agent_permissions(
|
|||
&format!("no hive's wanted state declares an agent {agent:?} that is not destroyed"),
|
||||
)));
|
||||
}
|
||||
if let Some(perms) = &req.permissions
|
||||
&& grant_of(perms) != Grant::default()
|
||||
{
|
||||
let (Some(wanted), hives) = (state.wanted.as_ref(), hive_names(&state)) else {
|
||||
return Err(refuse(error_problem(
|
||||
StatusCode::SERVICE_UNAVAILABLE,
|
||||
"no swarm queue is configured on this host",
|
||||
)));
|
||||
};
|
||||
require_tolerant_hive(&client, wanted, &hives, &agent)
|
||||
.await
|
||||
.map_err(|why| refuse(error_problem(StatusCode::CONFLICT, &why)))?;
|
||||
}
|
||||
let store = open_or_create(&client).await?;
|
||||
for _ in 0..MAX_ATTEMPTS {
|
||||
let (current, revision, undecodable) = read_or_default::<AgentPermissions>(&store, &agent)
|
||||
|
|
@ -200,6 +240,7 @@ pub(crate) async fn put_agent_permissions(
|
|||
.map_err(|e| queue_problem(&e))?
|
||||
{
|
||||
tracing::info!(%agent, ?value, "set agent permissions");
|
||||
mirrored(&agent, ¤t, &value).await?;
|
||||
return Ok(Json(view(value)));
|
||||
}
|
||||
}
|
||||
|
|
@ -255,6 +296,7 @@ pub(crate) async fn delete_agent_permissions(
|
|||
continue;
|
||||
}
|
||||
tracing::info!(%agent, "cleared agent permissions");
|
||||
mirrored(&agent, ¤t, &kept).await?;
|
||||
return Ok(StatusCode::NO_CONTENT);
|
||||
}
|
||||
Err(queue_problem(&anyhow::anyhow!(
|
||||
|
|
@ -271,10 +313,17 @@ pub(crate) async fn delete_agent_permissions(
|
|||
pub(crate) async fn clear_on_destroy(state: &AppState, hive: &str, agent: &str) {
|
||||
let cleared = async {
|
||||
let client = queue(state).map_err(|(_, why)| anyhow::anyhow!(why))?;
|
||||
agent_permissions::open_or_create(&client)
|
||||
.await?
|
||||
.delete(agent)
|
||||
.await?;
|
||||
let store = agent_permissions::open_or_create(&client).await?;
|
||||
let (current, _, _) = read_or_default::<AgentPermissions>(&store, agent).await?;
|
||||
store.delete(agent).await?;
|
||||
// The mirrored grant goes with the record.
|
||||
mirror(
|
||||
&observability_grant::Store,
|
||||
agent,
|
||||
¤t.permissions.unwrap_or_default(),
|
||||
&[],
|
||||
)
|
||||
.await?;
|
||||
anyhow::Ok(())
|
||||
}
|
||||
.await;
|
||||
|
|
@ -457,10 +506,15 @@ pub(crate) fn spawn(client: async_nats::Client, wanted: Arc<WantedWriter>, hives
|
|||
return;
|
||||
};
|
||||
tokio::spawn({
|
||||
let (store, wanted, hives) = (store.clone(), Arc::clone(&wanted), hives.clone());
|
||||
let (client, store, wanted, hives) = (
|
||||
client.clone(),
|
||||
store.clone(),
|
||||
Arc::clone(&wanted),
|
||||
hives.clone(),
|
||||
);
|
||||
async move {
|
||||
// Retries without end, so it answers only once it has worked.
|
||||
let _ = (|| seed_ruth_grant(&store, &wanted, &hives))
|
||||
let _ = (|| seed_ruth_grant(&client, &store, &wanted, &hives))
|
||||
.retry(every_five_seconds())
|
||||
.notify(|e: &anyhow::Error, _| {
|
||||
tracing::warn!(
|
||||
|
|
@ -539,6 +593,7 @@ const SEEDED_AGENT: &str = "ruth";
|
|||
/// `ruth` whose permissions were unset; this writes that grant into her key
|
||||
/// so she keeps her recovery mounts. Idempotent: a set key is never touched.
|
||||
async fn seed_ruth_grant(
|
||||
client: &async_nats::Client,
|
||||
store: &async_nats::jetstream::kv::Store,
|
||||
wanted: &WantedWriter,
|
||||
hives: &[String],
|
||||
|
|
@ -549,16 +604,27 @@ async fn seed_ruth_grant(
|
|||
declarations.push(d);
|
||||
}
|
||||
}
|
||||
seed_into(store, &crate::agent_renewal::live_agents(&declarations)).await?;
|
||||
Ok(())
|
||||
let declared = crate::agent_renewal::live_agents(&declarations);
|
||||
if !declared.contains(SEEDED_AGENT) {
|
||||
return Ok(());
|
||||
}
|
||||
let tolerant =
|
||||
observability_grant::hive_is_tolerant(client, wanted, hives, SEEDED_AGENT).await?;
|
||||
seed_into(store, &declared, tolerant).await?;
|
||||
// At every start, so a mirror that failed to write is retried.
|
||||
let (current, _, _) = read_or_default::<AgentPermissions>(store, SEEDED_AGENT).await?;
|
||||
let held = current.permissions.unwrap_or_default();
|
||||
mirror(&observability_grant::Store, SEEDED_AGENT, &[], &held).await
|
||||
}
|
||||
|
||||
/// Write [`Permission::DEFAULT`] plus `manage_root_agent` under
|
||||
/// [`SEEDED_AGENT`]'s key when `declared` holds her and her `permissions`
|
||||
/// are unset (no key, a deleted key, or `{}`). `true` when it wrote.
|
||||
/// Write [`Permission::DEFAULT`] plus `manage_root_agent`, and `logs` and
|
||||
/// `metrics` when her hive is `tolerant`, under [`SEEDED_AGENT`]'s key when
|
||||
/// `declared` holds her and her `permissions` are unset (no key, a deleted
|
||||
/// key, or `{}`). `true` when it wrote.
|
||||
async fn seed_into(
|
||||
store: &async_nats::jetstream::kv::Store,
|
||||
declared: &BTreeSet<String>,
|
||||
tolerant: bool,
|
||||
) -> anyhow::Result<bool> {
|
||||
if !declared.contains(SEEDED_AGENT) {
|
||||
return Ok(false);
|
||||
|
|
@ -570,6 +636,9 @@ async fn seed_into(
|
|||
}
|
||||
let mut perms = Permission::DEFAULT.to_vec();
|
||||
perms.push(Permission::ManageRootAgent);
|
||||
if tolerant {
|
||||
perms.extend([Permission::Logs, Permission::Metrics]);
|
||||
}
|
||||
let value = AgentPermissions {
|
||||
permissions: Some(perms),
|
||||
..AgentPermissions::default()
|
||||
|
|
@ -616,7 +685,7 @@ async fn import_into(
|
|||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
pub(crate) mod tests {
|
||||
use super::{
|
||||
Permissions, delete_agent_permissions, get_agent_permissions, import_into,
|
||||
put_agent_permissions, seed_into,
|
||||
|
|
@ -701,7 +770,7 @@ mod tests {
|
|||
);
|
||||
}
|
||||
|
||||
async fn published(client: &async_nats::Client, agent: &str) -> Option<String> {
|
||||
pub(crate) async fn published(client: &async_nats::Client, agent: &str) -> Option<String> {
|
||||
let store = swarm_queue_client::agent_permissions::open_or_create(client)
|
||||
.await
|
||||
.expect("bucket");
|
||||
|
|
@ -945,6 +1014,91 @@ mod tests {
|
|||
assert_eq!(published(&client, "atlas").await.as_deref(), Some(older));
|
||||
}
|
||||
|
||||
async fn publish_snapshot(client: &async_nats::Client, json: &str) {
|
||||
swarm_queue_client::status::open_or_create(client)
|
||||
.await
|
||||
.expect("bucket")
|
||||
.put("pr1ma", json.as_bytes().to_vec().into())
|
||||
.await
|
||||
.expect("snapshot");
|
||||
}
|
||||
|
||||
fn observing(perms: &[Permission]) -> Permissions {
|
||||
Permissions {
|
||||
permissions: Some(perms.to_vec()),
|
||||
unrestricted_internet: None,
|
||||
internet_log: None,
|
||||
}
|
||||
}
|
||||
|
||||
/// `logs` and `metrics` are refused, naming the hive and writing nothing,
|
||||
/// while the agent's hive has published no snapshot or one without the
|
||||
/// capability. Control: a grant that names neither passes the same hive.
|
||||
#[tokio::test]
|
||||
async fn logs_and_metrics_need_a_hive_that_advertises_tolerant_decoding() {
|
||||
let nats = TestNats::start();
|
||||
let client = nats.client().await;
|
||||
let state = crate::agent_limits::tests::declared(&client).await;
|
||||
|
||||
for perm in [Permission::Logs, Permission::Metrics] {
|
||||
let none = put(&state, "atlas", observing(&[perm]))
|
||||
.await
|
||||
.expect_err("no snapshot");
|
||||
assert_eq!(none.status, Some(StatusCode::CONFLICT));
|
||||
assert!(none.detail.as_deref().unwrap_or_default().contains("pr1ma"));
|
||||
}
|
||||
|
||||
publish_snapshot(&client, r#"{"status":"ok","warnings":[]}"#).await;
|
||||
let old = put(&state, "atlas", observing(&[Permission::Logs]))
|
||||
.await
|
||||
.expect_err("old hive");
|
||||
assert_eq!(old.status, Some(StatusCode::CONFLICT));
|
||||
assert_eq!(published(&client, "atlas").await, None, "nothing written");
|
||||
|
||||
let _ = put(&state, "atlas", observing(&[Permission::Messaging]))
|
||||
.await
|
||||
.expect("control: no log or metric grant");
|
||||
}
|
||||
|
||||
/// With the capability advertised the gate passes: the record is written,
|
||||
/// and the mirror (which needs the secret store, absent here) is what
|
||||
/// fails.
|
||||
#[tokio::test]
|
||||
async fn an_advertising_hive_passes_the_gate() {
|
||||
let nats = TestNats::start();
|
||||
let client = nats.client().await;
|
||||
let state = crate::agent_limits::tests::declared(&client).await;
|
||||
publish_snapshot(&client, r#"{"capabilities":["tolerant-permissions"]}"#).await;
|
||||
|
||||
let problem = put(&state, "atlas", observing(&[Permission::Metrics]))
|
||||
.await
|
||||
.expect_err("no secret store to mirror to");
|
||||
assert_eq!(problem.status, Some(StatusCode::INTERNAL_SERVER_ERROR));
|
||||
assert_eq!(
|
||||
published(&client, "atlas").await.as_deref(),
|
||||
Some(r#"{"permissions":["metrics"]}"#)
|
||||
);
|
||||
}
|
||||
|
||||
/// With her hive tolerant ruth is seeded with `logs` and `metrics`; with
|
||||
/// an old hive she gets the set she always had.
|
||||
#[tokio::test]
|
||||
async fn the_seed_includes_logs_and_metrics_on_a_tolerant_hive() {
|
||||
let nats = TestNats::start();
|
||||
let client = nats.client().await;
|
||||
let store = swarm_queue_client::agent_permissions::open_or_create(&client)
|
||||
.await
|
||||
.expect("bucket");
|
||||
let declared = ["ruth".to_owned()].into();
|
||||
assert!(seed_into(&store, &declared, true).await.expect("seeded"));
|
||||
assert_eq!(
|
||||
published(&client, "ruth").await.as_deref(),
|
||||
Some(
|
||||
r#"{"permissions":["messaging","meta","inbox","execution","manage_root_agent","logs","metrics"]}"#
|
||||
)
|
||||
);
|
||||
}
|
||||
|
||||
/// A name no hive declares, or a name that is not an identifier, is a 400
|
||||
/// and writes nothing.
|
||||
#[tokio::test]
|
||||
|
|
@ -1170,25 +1324,29 @@ mod tests {
|
|||
.expect("bucket");
|
||||
let declared = ["atlas".to_owned(), "ruth".to_owned()].into();
|
||||
|
||||
assert!(seed_into(&store, &declared).await.expect("seeded"));
|
||||
assert!(seed_into(&store, &declared, false).await.expect("seeded"));
|
||||
assert_eq!(published(&client, "ruth").await.as_deref(), Some(SEEDED));
|
||||
assert_eq!(published(&client, "atlas").await, None, "only ruth");
|
||||
assert!(!seed_into(&store, &declared).await.expect("idempotent"));
|
||||
assert!(
|
||||
!seed_into(&store, &declared, false)
|
||||
.await
|
||||
.expect("idempotent")
|
||||
);
|
||||
|
||||
store.delete("ruth").await.expect("delete");
|
||||
assert!(seed_into(&store, &declared).await.expect("seeded"));
|
||||
assert!(seed_into(&store, &declared, false).await.expect("seeded"));
|
||||
assert_eq!(published(&client, "ruth").await.as_deref(), Some(SEEDED));
|
||||
|
||||
store
|
||||
.put("ruth", b"{}".to_vec().into())
|
||||
.await
|
||||
.expect("unset");
|
||||
assert!(seed_into(&store, &declared).await.expect("seeded"));
|
||||
assert!(seed_into(&store, &declared, false).await.expect("seeded"));
|
||||
assert_eq!(published(&client, "ruth").await.as_deref(), Some(SEEDED));
|
||||
}
|
||||
|
||||
/// A ruth key the swarm already set, even to nothing, is left as it is;
|
||||
/// so is one this build cannot decode.
|
||||
/// so is one holding only a name this build does not know.
|
||||
#[tokio::test]
|
||||
async fn the_seed_leaves_a_set_ruth_key_alone() {
|
||||
let nats = TestNats::start();
|
||||
|
|
@ -1206,7 +1364,10 @@ mod tests {
|
|||
.put("ruth", set.as_bytes().to_vec().into())
|
||||
.await
|
||||
.expect("set");
|
||||
assert!(!seed_into(&store, &declared).await.expect("read"), "{set}");
|
||||
assert!(
|
||||
!seed_into(&store, &declared, false).await.expect("read"),
|
||||
"{set}"
|
||||
);
|
||||
assert_eq!(published(&client, "ruth").await.as_deref(), Some(set));
|
||||
}
|
||||
}
|
||||
|
|
@ -1222,12 +1383,12 @@ mod tests {
|
|||
.expect("bucket");
|
||||
|
||||
let declared = ["atlas".to_owned()].into();
|
||||
assert!(!seed_into(&store, &declared).await.expect("read"));
|
||||
assert!(!seed_into(&store, &declared, false).await.expect("read"));
|
||||
assert_eq!(published(&client, "ruth").await, None);
|
||||
assert_eq!(published(&client, "atlas").await, None);
|
||||
|
||||
let declared = ["atlas".to_owned(), "ruth".to_owned()].into();
|
||||
assert!(seed_into(&store, &declared).await.expect("seeded"));
|
||||
assert!(seed_into(&store, &declared, false).await.expect("seeded"));
|
||||
assert_eq!(published(&client, "ruth").await.as_deref(), Some(SEEDED));
|
||||
}
|
||||
|
||||
|
|
|
|||
Loading…
Reference in a new issue